Inputs to Motor Cortex

Introduction

Primary motor cortex (M1) doesn't work alone; it receives inputs from various sources to form motor commands.

Somatosensory Afferent Input to M1

Many M1 neurons have somatosensory receptive fields, like those in the somatosensory cortex.
Figure 1 shows somatosensory receptive fields in M1 of a monkey, identified through electrode tracks.
Sensory receptive fields were identified at certain locations.
Some responded to deep pressure on the hand (circle with an X).
Others responded to tactile stimuli like brushing the skin or passive extension of the index finger (blackened area, curved arrow).
Some sites responded to deep pressure and extension of all fingers, tactile stimuli of distal segments of the ring and little fingers, or extension of the ring and little fingers.
Finger extension responses likely come from sensory receptors stretched by the stimulus, like muscle spindles in finger flexor muscles.
Deep pressure responses likely activated receptors in finger flexor muscles whose tendons pass through the palm.
Electrical stimulation along these pathways mainly caused flexion or extension of fingers.
Sensory inputs to M1 were coherent with the actions caused by output from that region.
Lemon and Porter (1976) found that 210 out of 257 neurons tested had somatosensory receptive fields arising from the contralateral arm.
Most responded to movements in one direction of the shoulder, elbow, wrist, or fingers, or palpation of muscles, likely from proprioceptors.
Only 10% responded to tactile stimulation (light touch) and generally had small receptive fields.
M1 receives rich somatosensory input, providing detailed information about body part configuration and movements, which M1 needs to organize commands associated with voluntary motor behaviors.
Somatosensory input to M1 comes either from the somatosensory cortex or directly from somatosensory afferents via the dorsal column pathway and thalamus (see Figure 1, Chapter 11).
M1 receives input from the thalamus, especially from the cerebellum and basal ganglia.
Significant input to M1 comes directly from the somatosensory thalamus.
Thalamic neurons receive short latency excitatory inputs from somatosensory afferents (tactile and proprioceptive).
These neurons are excited antidromically by stimulation in M1, indicating that their axons project into M1.
These neurons aren't driven antidromically by stimulating in primary somatosensory cortex (S1).
Two separate populations of neurons in the thalamus transmit information from tactile and proprioceptive afferents to M1: one indirectly through S1 and another directly to M1 (Figure 2).
Recent findings confirm direct inputs to the motor cortex from the sensory thalamus.

Cortical Inputs to M1

M1 receives inputs from other cortical regions.
Figure 3 shows the main cortical connections into M1.
One source of input to M1 is from the primary somatosensory cortex (S1).
S1 input provides additional and more highly processed somatosensory information than that supplied more directly from the thalamus (Figure 2).
Little of the input from S1 derives from the tactile region Brodmann’s area 3b.
Brodmann’s area 1, processes tactile signals and has extensive inputs to M1, providing cortically processed tactile information.
The most extensive inputs to M1 arise from Brodmann’s area 6, called the premotor cortex.
The premotor cortex has two subdivisions, a dorsal medial region called the supplemental motor area (SMA) and a lateral region called the lateral premotor cortex (LPMC) (Figure 3).
These regions are involved in organizing plans for impending voluntary movements, including which muscles to activate and when to achieve a desired movement goal.
The premotor areas are the architects of voluntary movements, and M1 is the construction company executing the plans.
SMA plans internally generated movements, while LPMC plans externally guided movements.
Internally generated movements don’t involve an overt physical target (e.g., waving hand around in space, playing a memorized piece of music).
Externally guided movements are guided to an external signal or object (e.g., hitting a tennis ball, reaching for a coffee cup).
Formulating a movement plan requires critical information to be provided to the premotor areas.
When reaching for a coffee cup (Figure 4A), one needs the location of the cup (x, y, z{\text{cup}}), which is derived from the visual system, and the limb orientation/hand position (x, y, z{\text{hand}}), which is identified through proprioceptive information.
The LPMC can prepare a reach plan involving particular muscles activated at various intensities to drive the hand to the cup.
If the cup stays in place but the arm is configured differently (Figure 4B), a different reach plan is devised based on new proprioceptive information.
Brodmann’s area 7, a high-order association cortex, is part of the dorsal stream of visual processing and provides rich inputs to area 6, encoding spatial information from visual signals, identifying the relative locations of objects in a visual field.
The ventral stream of visual processing extracts the identity of objects in the visual field.
These two streams are the “where” and “what” pathways of the visual system.
Area 7 encodes locations of objects within reaching distance and of interest to grasp.
Quote from Mountcastle and colleagues (1975):
- "The discharge of [area 7] cells increase abruptly when the animal [visually] fixates certain objects. The object must, to be effective, be of interest to the animal, such as food. The effect of such an object is maximal if it is located within arm’s reach, and the associated discharge on fixation decreases with object distance . . . Such a cell may, for example, be completely silent as the animal explores his environment visually fixating in sequence experimenter, instruments, books, etc., to be followed immediately by intense discharge on fixation of a food object presented within arm’s length. That discharge increases as the target is moved closer to the animal."
Projections from area 7 into the LPMC provide spatial information about target objects.
Brodmann’s area 5 integrates information from the primary somatosensory cortices to generate a coded representation of overall limb orientation, which is then relayed to the LPMC and SMA.
Based on object location (from area 7) and arm/hand orientation (from area 5), the premotor cortex puts together a movement plan, forwarded to M1, which enacts the plan through commands sent to interneurons and motor neurons in the brainstem and spinal cord (Figure 3).

Supplementary Motor Area

SMA is involved in planning internally generated movements.
Experiments involving brain activity and blood flow were used to pinpoint the SMA.
Increased blood flow was detected using radioactivity detectors surrounding the skull after injecting a short-acting radioactive substance.
Subjects squeezed and released a spring between their thumb and index finger (Figure 5A), resulting in increased blood flow in M1 and S1 (Figure 5B).
Subjects learned a 16-step sequence of finger taps (Figure 5C), resulting in increased blood flow in M1, S1, and the dorsal, medial region of Brodmann’s area 6 (SMA) (Figure 5D).
Because the remembered sequence represents an internally generated behavior, SMA was implicated in the production of such behaviors.
Subjects mentally rehearsed the memorized sequence without moving the fingers (Figure 5E), activating only the SMA (Figure 5F).
This indicates that the SMA is involved in the planning but not necessarily the execution of internally generated movements.

Lateral Premotor Cortex

LPMC is involved in planning movements associated with externally guided behaviors, like reaching for an object.
Studies with monkeys trained to wait for a go cue before reaching to a displayed target supported this idea (Figure 6).
Monkeys reached to one of three or four targets as they were illuminated, keeping their hand on the target until instructed to reach the next target.
Even though the next target was illuminated, the monkey waited until a “go-cue” was turned on.
Figure 6A shows the spiking activity of a neuron recorded in the dorsal region of the LPMC during trials where the hand was on the third target from the right and the upcoming target was immediately to the left.
- The neuron started firing when the target was illuminated and maintained activity throughout the wait period, dissipating once the go-cue was displayed and the monkey reached the target.
Figure 6B shows activity of the same neuron when it reaches to the right.
- There was little activity upon display of the target, but activity increased after the monkey reached the rightmost target, as if preparing for upcoming movements.
This particular neuron was involved in the planning of leftward movements to targets.
Thousands of neurons are involved in preparation for upcoming movements to visual targets, tuned to different movement directions.

Concurrent Activities in SMA, LPMC, and M1

Experiments with monkeys carrying out sequential reaching movements to targets under two conditions further distinguished SMA and LPMC (Figure 7).
In one, monkeys reached to and touched targets illuminated in a random sequence (Figure 7A), planning based on external, visual targets.
In the other, monkeys carried out a memorized, learned sequence of reaches to three targets (Figure 7B), triggering the learned sequence with a go-cue, using internally generated movement plans.
Neural activity could be compared for these two conditions because the movements were the same with only the planning contingencies different (external vs. internally generated cues).
Neural activity was recorded from M1, SMA, and LPMC for both conditions.
Figure 7C shows example neurons recorded from each area during the visually cued and internally cued conditions.
The M1 neuron (left column, Figure 7C) had similar activity for both conditions, with three bursts, one for each target.
The LPMC neuron (middle column, Figure 7C) was strongly activated only during the visually guided movements but not during the internally cued movements.
The SMA neuron (right column, Figure 7C) was little activated during the visually guided reaches while more robustly engaged during the internally cued movements.
LPMC is involved in planning externally guided movements, SMA in planning internally generated movements, and M1 in executing the plans.

Distinct Planning for Reaching and Grasping

Reaching behaviors involve both transportation (propelling the hand to the target object) and manipulation (configuring the hand and fingers to interact with the object) functions.
Planning transportation depends on the location of the hand and the target (Figure 4), involving Brodmann’s area 5 to identify hand position based on proprioceptive signals and Brodmann’s area 7 to register the location of the target based on visual information.
Successful grasping and manipulation require knowing the current configuration of the hand and fingers, derived from proprioceptive information.
A different subregion of area 5 encodes the spatial dispositions of the hand and fingers.
Planning for manipulating an object depends on the physical geometry and characteristics of the target object.
This data is extracted from visual information in the intraparietal sulcus, which separates area 5 from area 7.
Figure 8 shows recordings from a neuron in the anterior intraparietal (AIP) area in a monkey viewing different objects (Figures 8A–F).
- The objects were placed in the same position, and in some trials, monkeys reached and grabbed the objects.
- In these trials, the monkey stared at the object but did not reach for it.
- The neuron’s activity was strong when fixating on the plate (Figure 8A), moderately active when fixating on the ring or small cube (Figures 8B, 8C), and little active for the other objects.
- These patterns of activity were maintained throughout the reach when the monkey was allowed to grab the objects.
- Neurons in the AIP represent the spatial configuration and orientation of objects that might be grasped.
Projections from AIP go to the ventral region of LPMC, PMV.
Neurons in this region are involved in setting up the appropriate plan to grasp objects of various shapes and dimensions.
Figure 9 shows a recording of a neuron recorded in monkey PMV.
- The monkey sat in front of a box containing a turntable divided into sectors, each holding a different-shaped object.
- The objects were presented to the monkey in random order by rotation of the turntable and illuminated (small vertical arrows in Figure 9).
- Monkeys waited 1.5 seconds after seeing the object before a go-cue instructed them to reach and grab the object.
- Little activity was seen in this neuron for most of the objects, but the neuron was strongly activated upon visual illumination of the ring (Figure 9), and this was maintained until just after the go-cue was displayed. Some modest activity also occurred upon presentation of the sphere.
- Neurons in PMv showed selective activity during the wait period for different objects.
- This preparatory activity is similar to the activity of the LPMC neuron in Figure 6 in advance of reaching to targets in different locations.
- The neuron was recorded in the dorsal region of the LPMC, PMd.
PMd appears to be involved in planning of reaching, while PMv seems to plan how to move the fingers and hand to interact with objects in the environment.

Mirror Neurons

Some neurons recorded in PMv respond when a monkey grasps a particular object and when the monkey observes another monkey or human grasping the same object.
Figure 10 depicts the activity of such a mirror neuron.
In Figure 10A, the monkey watched a human experimenter pick up a morsel of food, and the monkey also picked up a food morsel.
- The neuron was robustly activated both by observing the action of others (“monkey see”) and performing the action himself (“monkey do”).
This mirroring of activity only occurred if the watched action was similar to the specific action that engaged the neuron prior to an actual movement.
Figure 10B shows the responses of the same neuron when the experimenter picked up the morsel with a pair of pliers.
- There was very little activity in the neuron during the observation phase, but strong activity occurred when the monkey picked up the morsel with his fingers.
There has been considerable debate as to the meaning of mirror neuron activity.
It might represent a kind of inadvertent mental rehearsal associated with planning to perform the movement, triggered by the observation of someone else doing it.
It has also been suggested that such mirror neuron activity might underlie the learning of new motor behaviors based on observation of others.
Mirror neurons have been suggested to play a key role in social cognition.
- The ability to understand the intentions, actions, or emotions of others partly derives from the activity of mirror neurons that “simulate,” in the observer’s mind, the neural activity that would occur if the observer were to be involved in the same action.
- When someone observes someone weeping, activation of mirror neurons associated with that behavior might impart a feeling of sadness, even if the observer does not know why the other person is crying.
The function of mimicking activity by mirror neurons is not yet fully understood.

Neural Population Dynamics

A deeper understanding of how neural activity associated with movement planning leads to a particular pattern of descending motor commands requires consideration of representative populations of the tens of thousands of neurons simultaneously involved.
A new conceptual framework, neural population dynamics, is used to gain insight into how the collective actions of many neurons during the planning stages of movement give rise to the complex motor commands that drive muscles during voluntary behaviors.
The concept of neural population dynamics derives from the field of dynamical systems, used in mathematics and physics to describe systems whose properties change over time.
Insight into the behavior of a dynamical system is gained by graphical depiction of key variables that characterize the system.
For a pendulum (Figure 11A), the horizontal axis might indicate how far to the right or to the left of the straight down position the pendulum is at any moment (Figure 11B), while the vertical axis represents the velocity of the pendulum (with rightward velocities indicated as positive and leftward as negative, Figure 11B).
The plot will have a circular-shaped trajectory: when the pendulum is at its far-left position, its velocity is zero (position 1, Figure 11). Then, as it starts to move to the right, the velocity increases, reaching a maximum as it passes through the zero position (position 2, Figure 11). Then the velocity progressively slows down until it again reaches zero at the far-right position (position 3, Figure 11), and so on.
Any point on this trajectory indicates the state (position, velocity) of the pendulum at that moment and is predictive of forthcoming states.
In neural population dynamics, the key variables representing the brain system of interest are the firing rates of individual neurons, like position and velocity for a pendulum.
The collective spiking activity across a population can be plotted where each axis on the graph represents the firing rate of an individual neuron.
Figure 12A shows the spiking activity of three neurons. For each successive brief time period (shaded regions), the spiking rates of each neuron can be determined and plotted as points in 3D space (one dimension for each neuron, \text{n}1, \text{n}2, \text{n}_3, Figure 12B).
For example, for the first time period, neuron 1 generated zero spikes, neuron 2 four spikes, and neuron 3 zero spikes. The activities of these neurons at this time are then represented as the white dot in the 3D firing-rate space (sometimes called the neural state space) (Figure 12B).
Connecting the dots across each successive state produces a path through the neural state space called a neural trajectory (Figure 12B).
Representing the activities of multiple neurons allows clear visual depiction of how the activities of multiple neurons evolve over time.
Representing a neural population requires recording the activities of a few hundred neurons simultaneously.
Data scientists use “tricks” that maintain the richness of large-scale multidimensional data while simplifying the representation to a few visual dimensions, called dimensionality reduction.
This approach capitalizes on common (or correlated) behaviors among some neurons.
Figure 13 shows an example of this idea. The simultaneous spiking activities of two neurons (\text{n}1, \text{n}2) are illustrated in Figure 13A. To represent the neural dynamics of this population of two neurons requires a two-dimensional state space (Figure 13B).
The firing rates of the two neurons at different epochs of time (1, 2, 3, and 4, Figure 13A) are shown as points in this space (Figure 13B).
These points are constrained to reside along a single line. The firing rates of the two neurons appear to go up and down together, as if the activity of one neuron was harnessed to the activity of the other.
Two separate dimensions are not required to represent the activities of these two neurons.
We can characterize their firing along a single tilted axis that passes through most of the data points (Figure 13C).
A method used to identify such axes that can account for much of the data along fewer axes than the original dimensions is called principal components analysis, with each new axis called a principal component (pc, Figure 13C).
We can envision the activities of the two neurons as moving together along a single pc axis (Figure 13D).
We have lessened the number of dimensions of the neural state space from two to one.
Neurons whose activities do not follow along such a path, or for more complex neural activity, would need additional pc dimensions.
A set of principal components might effectively represent activity among a group of neurons for one behavior or mental process, while a different set might be needed for other situations.
Figure 14A shows a neural trajectory. The population activity does not appear to visit all possible regions of the 3D space.
The population activity largely is confined to a single plane.
Because a plane can be defined using just two orthogonal axes (in this case, pc1 and pc2), we can illustrate the simultaneous activities of a 3-neuron population using two dimensions (Figure 14B).

Neural Population Dynamics During Movement Planning

Neural activity distributed over a population of neurons involved in planning a movement in premotor areas leads to a particular pattern of descending motor commands.
Krishna Shenoy, Mark Churchland, and their colleagues proposed that neural activity before movement (preparatory activity) positions the neural population at a particular site in neural state space, seeding and dictating the impending neural dynamics associated with movement execution.
Preparatory activity is like positioning a pendulum to some location prior to letting it go.
Consider a large population of premotor cortex neurons whose activities can be represented by three principal components.
Figure 15A shows activity in the neural state space associated with planning and making movements to a target on the right.
- Before the target display, the neurons discharge somewhat randomly at relatively low intensities, referred to as baseline activity.
- Once a right target has been displayed, preparatory activity arises, driving the neural population to a new state. Neural activity hovers at this preparatory state for rightward movements until a go-cue is given.
- A particular pattern of neural activity ensues (the swinging of the pendulum).
- This is registered as a trajectory through neural state space (movement activity, Figure 15A), which impels primary motor cortex to emit signals driving muscles to perform the desired rightward movement.
In the case of movements to a target on the left (Figure 15B), preparatory activity pushes the neural population to a different state that seeds neural activity associated with movement to the left.
If there were no requirement to wait for a go-cue, movement activity would immediately follow once preparatory activity attained the desired preparatory state.
Preparatory and movement activities can be envisioned as a person at a waterslide park (Figure 15C).
- Preparatory activity involves climbing stairs and ladders to a landing in front of a desired slide entrance.
- The person may be required to “hover” before the attendant gives the “go” signal.
- The ensuing movement activity is dictated by the physical characteristics and linkages among the structural components of that specific slide.
- Different slides will propel the individual to different locations in the pool.
- Preparatory activity situates the individual at a start point needed to seed a forthcoming dynamical ride on a desired slide.
An example of the preparatory activity of a neural population is shown in Figure 16.
Monkeys were trained to reach to one of eight targets from a center position displayed on a touch screen (Figure 16A).
Monkeys were required to wait for up to 1 second after the target was displayed before a go-cue instructed the monkey to reach the target.
Figure 16B shows the normalized firing rates of 127 neurons (recorded mainly in PMd) represented in neural state space with just two dimensions (PC1 and PC2) during the wait period prior to reaching the different targets.
- Preparatory activity propelled the neural population from baseline to one of eight distinct preparatory states (arrowheads, Figure 16B) that were maintained until the go-cue was presented.
- Following the go-cue, a unique neural trajectory emerged from each state associated with movement execution to each of the targets.
- The locations of the preparatory states mimic the spatial arrangements of the targets in physical space.

Summary

The primary motor cortex depends on sensory input and preparatory information from other parts of the brain to generate movement commands.
During reaching to an object (Figure 17), the image of the object is registered on the retinas (a) and conveyed (b) through the visual thalamus to the primary visual cortex, V1.
Simultaneously, proprioceptors in the arm and hand register their orientations (c) and convey that information (d) through a different region of the thalamus to the primary somatosensory cortex, S1.
Visual information about the location of the cup and its physical dimensions are processed along the dorsal visual stream (e).
In Brodmann’s area 7, the location of the object with respect to the body is encoded, whereas the spatial configuration of the object to be grasped is represented in the AIP cortex.
Similarly, information from the various subdivisions of S1 converge in Brodmann’s area 5 (f) to holistically represent the position of the hand and configuration of the fingers.
Signals from areas 7 (g) and 5 (h), representing target and hand locations, are passed onto the planning region for arm movements, the dorsal premotor cortex (PMd).
Likewise, information about finger configurations and object attributes is forwarded from area 5 (i) and AIP (j) to the planning region for grasping actions, the ventral premotor cortex (PMv).
The plans for moving the arm (k) and fingers (l) are relayed to the primary motor cortex (M1), perhaps by seeding particular preparatory states that dictate the neural dynamics underlying descending commands (m).
Those commands operate on interneurons and motor neurons in the brainstem and spinal cord to engage muscles in the arm and hand (n).
Finally, that muscular activity transports the hand (o) to the vicinity of the desired object and arranges the digits of the hand (p) appropriately for grasping the object.
If no overt external target were the goal of the movement, then the SMA would provide the planning information conveyed to M1 to execute such internally generated movements.